"March 2009"--cover date

Includes bibliographical references (p. 22-23)

Wetlands are increasingly at risk from human alteration of the landscape. Although site-specific activities like have the most direct and obvious impacts on wetland integrity, activities within the surrounding catchment can also lead to degradation by changing wetland hydrologic function, increasing nutrient and sediment loads, and providing a conduit for the spread of invasive and exotic species. With the widespread adoption of GIS technology, it has become possible to characterize large landscapes and identify potential stressors from existing datasets. Because so much information is available on a desktop computer, the U.S. Environmental Protection Agency advocates the use of GIS-based landscape analysis to provide a preliminary assessment of wetland condition in a project area (Level I), before conducting field-based rapid (Level II) and intensive (Level III) assessments. Although most Level I assessment approaches are developed with best professional judgment, when field data is available, it can support development, calibration and validation of metrics. In Montana, we have rapid assessment data on over a thousand wetlands across the state. Our goal in this study was to determine whether we could use this data to identify landscape-level metrics with a good ability to predict wetland condition, or, at the least, to calibrate and validate a best professional judgment-based tool. From a review of the literature, we identified a number of landscape-scale metrics that are widely believed to influence wetland condition. We calculated values for these metrics in several different buffer distances for a random sample of 591 wetlands, and performed several statistical analyses (ANOVA, stepwise regression, CART) to find metrics with significant relationships to the field-determined overall condition scores. At the 6th code Hydrologic unit (HUC), 1 kilometer, 500 meter, and 200 meter buffer distance, the combined metrics of percent forest cover, road density, and number of stream road crossings had the strongest predictive value for overall score. We had observed that there was a strong ecoregional skew in the condition scores, with wetlands in mountain ecoregions having a higher average score than wetlands in plains ecoregions, so we split the assessment data into a mountain and a plains subsets and reran the analysis. With the data divided, percent forest was no longer significant at any scale. For wetlands in the mountain ecoregions (n=262), road density was the only metric that was significant at all levels, although the R-squared value was never higher than 0.07. In the 1 kilometer buffer, the percentage of crop agriculture was also significant, although it had no significance at other buffer distances. In the plains ecoregions, no metrics were significant at 200 meters. Percent natural grassland and road density within 500 meters were both significantly correlated with overall score but had very low R-squared value (0.02 and 0.01, respectively). At the 1,000 meter buffer scale, only the number of stream road crossings was significant. No metric was significantly correlated to overall wetland condition when measured at the 6th code HUC level in either the mountain dataset or the plains dataset. When we added an environmental variable (relative effective annual precipitation) to the analysis, we found it had high predictive value for the dataset as a whole, and within the subset of mountain ecoregions. In the plains, where it varied less, it was not significant. Using best professional judgment, we then built a Montana Landscape Integrity Model (MT-LIM) and used the dataset to calibrate it. The model is an inverse weighted distance model premised on the idea that ecosystem processes and functions achieve their fullest expression in areas where human activities have the least impact. The model was used to calculate a mean landscape integrity score for pixels within 100 meters of a wetland. This score was combined with a relative effective precipitation value from the assessment point, and wetlands were assigned to an ordinal condition class (A, B, C or D) using thresholds we identified through calibration. When compared to the condition classes assessed in the field using rapid assessment methods, this approach accurately predicted the measured condition in 50-55% of cases, depending on field method. In mountain ecoregions, it accurately predicted A-ranked wetlands in 75% of the cases. Sixty-five percent of the plains wetlands and 83.5% of the mountain wetlands were classified acceptably, i.e., the classifications were either accurate or no more than one rank higher than what was assigned in the field. This study demonstrated the potential of landscape-level metrics and models to predict wetland condition using remotely-sensed data in Montana. At the same time, it showed that environmental variables and site-specific activities may be far more important drivers of wetland condition than land uses occurring at a broader scale

"This project was funded by an EPA Region 8 Wetland Program Development Grant, administered by the Montana Department of Environmental Quality."-p. v. Agreement Number

NB-MSL

plus-circle Add Review

comment

Reviews